Changkyo Lee

23.2 A 5Gb/s/pin 8Gb LPDDR4X SDRAM with power-isolated LVSTL and split-die architecture with 2-die ZQ calibration scheme

With growing demand for low-power mobile applications, such as wearable devices, smart phones and tablet PCs, low-power mobile DRAM has been identified as a mandatory requirement for low-power system designs. The recently developed LPDDR4 [1] is still a power efficient solution because of its architectural approaches and low-voltage-swing terminated logic (LVSTL). However, demand for enhanced power-efficiency beyond LPDDR4 is still increasing for mobile applications. In this work, a 5.0Gbp/s/pin 8Gb LPDDR4X memory with power-isolated low-voltage-swing terminated logic (PI-LVSTL) and a split-die architecture is proposed to enhance power-efficiency and mass production yield.

A 6.4Gb/s/pin at sub-1V supply voltage TX-interleaving technique for mobile DRAM interface

A 6.4Gb/s TX-interleaving (TI) technique at sub-1V supply voltage is implemented with 25nm DRAM process for the future mobile DRAM interface which requires 51.2 GBps (2X Bandwidth of LPDDR4). A newly proposed 2-channel TX interleaving technique with a bootstrapping switch can save power consumption drastically by eliminating repeaters, while operating at 6.4 Gb/s with 40 % enhancement of I/O power efficiency compared to that of the LPDDR4.

Real-time 3D human pose recognition from reconstructed volume via voxel classifiers

This paper presents a human pose recognition method which simultaneously reconstructs a human volume based on ensemble of voxel classifiers from a single depth image in real-time. The human pose recognition is a difficult task since a single depth camera can capture only visible surfaces of a human body. In order to recognize invisible (self-occluded) surfaces of a human body, the proposed algorithm employs voxel classifiers trained with multi-layered synthetic voxels. Specifically, ray-casting onto a volumetric human model generates a synthetic voxel, where voxel consists of a 3D position and ID corresponding to the body part. The synthesized volumetric data which contain both visible and invisible body voxels are utilized to train the voxel classifiers. As a result, the voxel classifiers not only identify the visible voxels but also reconstruct the 3D positions and the IDs of the invisible voxels. The experimental results show improved performance on estimating the human poses due to the capability of inferring the invisible human body voxels. It is expected that the proposed algorithm can be applied to many fields such as telepresence, gaming, virtual fitting, wellness business, and real 3D contents control on real 3D displays.